Download - Phonons In The Elk Code
Phonons In The Elk Code
Kay Dewhurst, Sangeeta Sharma, Antonio Sanna,Hardy Gross
Max-Planck-Institut fur Mikrostrukturphysik, Halle
“If you are allowed to measure only one property of amaterial, make sure it’s the - - - - - - - - ”
A. Einstein
“If you are allowed to measure only one property of amaterial, make sure it’s the heat capacity”
A. Einstein
Why phonons?
I Measured directly with inelastic neutron scattering
I Thermodynamic quantities at low temperature: heatcapacity, entropy, free energy, zero-point
I Phase transitions from the Gibb’s free energy:
G(P, T ) = E + PV − TS
I Imaginary phonon frequencies indicate lattice instabilities
I Elastic tensor from q → 0 limit of phonon dispersion
I Thermal expansion coefficient (from mode Gruneisenparameters)
I Temperature dependence of the band gap
I Static polarization ε(ω = 0)
I Electron-phonon coupling
I Superconductivity
The ‘equilibrium positions’ of out crystal are the minimum ofthe Born-oppenheimer potential energy surface:
(Te + Ve−n(R) + Ve−e + En−n(R)
)ΦR(r) = E(R)ΦR(r)
This is an artificial construct and in some cases may be a poorapproximation to the average nuclear positions obtained fromthe correlated electron-nuclear wavefunction Ψ(r,R)
Recent work entitled ‘Exact Factorization of theTime-Dependent Electron-Nuclear Wave Function’ [A. Abedi,N. T. Maitra, and E. K. U. Gross, Phys. Rev. Lett. 105,123002 (2010)] may make it easier to handle these cases
Let’s assume that the potential energy surface is harmonic:
E(R) = UtCU + E(R0),
where U ≡ R−R0
Elastic matrix
We have to determine the spring constants between atoms:
Cαi,βj(R−R′) =∂2E
∂uαi(R)∂uβj(R′)
= − ∂Fαi(R)
∂uβj(R′)
for atoms α, β, polarisations i, j and primitive vectors R, R′
Thus
E = E0 +1
2
∑αiRβjR′
Cαi,βj(R−R′)uαi(R)uβj(R′) + · · ·
(no linear term)
Equation of motion:
Mα∂2
∂t2uαi(R) = −
∑βjR′
Cαi,βj(R−R′)uβj(R′)
Use the anzatz
uαi(R) =1√
2Mα
[eαi(q) exp(i(q ·R− ω(q)t)) + c.c.
]to get
Dαi,βj(q)eαi(q) = ω2(q)eαi(q)
where
Dαi,βj(q) =1√
MαMβ
∑R
exp(iq ·R)Cαi,βj(R)
Ab-initio approaches for computing the dynamical matrix
I Supercell Parlinski-Li-Kawazoe method: randomdisplacements and least-squares fitting (Wien2k, VASP)
I Supercell with molecular dynamics (Fourier transform oftrajectories)
I Supercell method used in Elk
I Density functional perturbation theory (DFPT)
Density functional perturbation theory (DFPT)
Start with the Hellmann-Feynman theorem
Let the Hamiltonian depend on some parameter λ, therefore
Eλ = 〈Ψλ|Hλ|Ψλ〉,
where Ψλ is the ground-state of Hλ
Take the derivative
∂Eλ∂λ
= 〈Ψλ|∂λHλ|Ψλ〉+ 〈∂λΨλ|Hλ|Ψλ〉+ 〈Ψλ|Hλ|∂λΨλ〉
= 〈Ψλ|∂λHλ|Ψλ〉+
Density functional perturbation theory (DFPT)
Start with the Hellmann-Feynman theorem
Let the Hamiltonian depend on some parameter λ, therefore
Eλ = 〈Ψλ|Hλ|Ψλ〉,
where Ψλ is the ground-state of Hλ
Take the derivative
∂Eλ∂λ
= 〈Ψλ|∂λHλ|Ψλ〉+ 〈∂λΨλ|Hλ|Ψλ〉+ 〈Ψλ|Hλ|∂λΨλ〉
= 〈Ψλ|∂λHλ|Ψλ〉+ ∂λ′ 〈Ψλ′ |Hλ|Ψλ′〉∣∣∣λ′=λ
Density functional perturbation theory (DFPT)
Start with the Hellmann-Feynman theorem
Let the Hamiltonian depend on some parameter λ, therefore
Eλ = 〈Ψλ|Hλ|Ψλ〉,
where Ψλ is the ground-state of Hλ
Take the derivative
∂Eλ∂λ
= 〈Ψλ|∂λHλ|Ψλ〉+ 〈∂λΨλ|Hλ|Ψλ〉+ 〈Ψλ|Hλ|∂λΨλ〉
= 〈Ψλ|∂λHλ|Ψλ〉+ ∂λ′ 〈Ψλ′ |Hλ|Ψλ′〉∣∣∣λ′=λ︸ ︷︷ ︸
= 0
Density functional perturbation theory (DFPT)
Only the external potential depends on λ
Hλ = T + V λext + Ve−e
Therefore
∂
∂λHλ =
∂
∂λV λext
⇒ ∂Eλ∂λ
= 〈Ψλ|∂λV λext|Ψλ〉
Thus finally
∂Eλ∂λ
=
∫d3rρλ(r)
∂
∂λV λext(r)
for all λ
Density functional perturbation theory (DFPT)
Taking another derivative:
∂2Eλ∂λ2
=
∫d3r
∂
∂λρλ(r)
∂
∂λV λext(r) +
∫d3r ρλ(r)
∂2
∂λ2V λext(r)
Difficult part:∂ρλ(r)
∂λ
Density functional perturbation theory (DFPT)
Recall the Kohn-Sham density is equal to the exact density
ρ(r) =∑i
φ∗i (r)φi(r)
So we get the linearised Kohn-Sham equations:
∆ei = 〈φi|∆Vs|φi〉(H − εi)|φi〉 = −(∆H −∆εi)|φi〉
∆ρ(r) =∑i
∆φ∗i (r)φi(r) + φ∗i (r)∆φi(r)
∆Vs(r) = ∆Vext(r) + ∆VH(r) +
∫d3r′
δExcδρ(r)δρ(r′)
∆ρ(r)
Density functional perturbation theory (DFPT)
If the perturbing potential is of the form
∆Vext(r) = − ∂
∂ταi
∑R
Zα|r + R− τα|
cos(q ·R)
then Bloch state |φik〉 is coupled to states |φik+q〉 and |φik−q〉the density derivative is
∆ρ(r) = ∆ρc(r) cos(q · r) + ∆ρs(r) sin(q · r)
and the same for sin(q ·R)
Density functional perturbation theory (DFPT)
We can also use perturbations of the form
∆Vext(r) = − ∂
∂ταi
∑R
Zα|r + R− τα|
exp(iq ·R)
in which case |φik〉 is coupled to |φik+q〉 only, and the densityderivative is
∆ρ(r) = ∆ρ(r) exp(iq · r)
and the effective potential derivative is also
∆Vs(r) = ∆Vs(r) exp(iq · r)
∆ρ and ∆Vs are lattice periodic
Density functional perturbation theory (DFPT)
This is very efficient, in the same way that spin-spirals areefficient for incommensurate magnetic structures
The APW basis set depends on the atomic positions – Pulaycorrections make implementation of phonons difficult
Previous implementations
I Henry Krakauer, Rici Yu and Cheng-Zhang Wang
I Robert Kouba and Claudia Ambrosch-Draxl in Wien2k
Supercell Linear response
Slow Fast
Easy to implement Hard to implementWorks with all features Works with fewer features
Does not require fxc Requires fxcCan include anharmonicity1 Purely harmonic
1Can be used for thermally stabilizing unstable phases: P. Souvatzis etal. 2009 J. Phys.: Condens. Matter 21 175402
Supercell Linear response
Slow FastEasy to implement Hard to implement
Works with all features Works with fewer featuresDoes not require fxc Requires fxc
Can include anharmonicity1 Purely harmonic
1Can be used for thermally stabilizing unstable phases: P. Souvatzis etal. 2009 J. Phys.: Condens. Matter 21 175402
Supercell Linear response
Slow FastEasy to implement Hard to implement
Works with all features Works with fewer features
Does not require fxc Requires fxcCan include anharmonicity1 Purely harmonic
1Can be used for thermally stabilizing unstable phases: P. Souvatzis etal. 2009 J. Phys.: Condens. Matter 21 175402
Supercell Linear response
Slow FastEasy to implement Hard to implement
Works with all features Works with fewer featuresDoes not require fxc Requires fxc
Can include anharmonicity1 Purely harmonic
1Can be used for thermally stabilizing unstable phases: P. Souvatzis etal. 2009 J. Phys.: Condens. Matter 21 175402
Supercell Linear response
Slow FastEasy to implement Hard to implement
Works with all features Works with fewer featuresDoes not require fxc Requires fxc
Can include anharmonicity1 Purely harmonic
1Can be used for thermally stabilizing unstable phases: P. Souvatzis etal. 2009 J. Phys.: Condens. Matter 21 175402
Phonons in Elk
Elk uses an efficient supercell method:
1. Finds the smallest supercell which contains the q-vectorperturbation
2. then performs calculations with cos-like and sin-like finitedisplacements
3. computes forces on all atoms and Fourier transforms togenerate dynamical matrix rows
4. gathers all the dynamical matrix rows together to producea phonon dispersion
Phonons in Elk
This can be run across hundreds of computers sharing acommon file system
Each computer checks for the presence of a particular DYN file
For example:DYN Q0104 0104 0102 S02 A002 P3.OUT
(means q = (14 ,14 ,
12), α = (1, 2), i = 3)
If it does not exist then the computer runs this task
Phonons in Elk
The phonon dispersion is obtained by Fourier transforming thedynamical matrices to real space:
Dαi, βj(R) =∑q
Dαi, βj(q) exp(iq ·R),
evaluating the matrix at an arbitrary q-vector:
Dαi, βj(q′) =∑R
Dαi, βj(R) exp(−iq ·R),
and diagonalising
The acoustic sum rule is always enforced (acoustic branches goto zero)
LO-TO splitting in polar semiconductors is not yet in Elk
Phonons in Elk
First-variational Hamiltonian (bordered band diagonal):
Phonons in Elk
First-variational Hamiltonian (bordered band diagonal):
Phonons in Elk
First-variational Hamiltonian (bordered band diagonal):
Elk also calculates
I Phonon density of states
I Heat capacity
I Entropy
I Free energy
I Zero point energy
I Electron-phonon coupling matrices
gnik+q,jk =1√
2Mωn(q)〈φik+q|en(q) ·∆Vs(q)|φjk〉
I Linewidths
γn(q) = 2πωn(q)|gnik+q,jk|2δ(εik+q − εF )δ(εjk − εF )
(interpolating like the dispersion)
I The Eliashberg function
α2F (ω) =1
2πNF
∑qn
γn(q)
ωn(q)δ(ω − ωn(q))
I Electron-phonon coupling parameter
λ = 2
∫ ∞0
dωα2F (ω)
ω
I McMillan-Allen-Dynes superconducting criticaltemperature
I Eliashberg equations and superconducting gap (AntonioSanna)
Phonon dispersion of bcc Nb (4× 4× 4 q-point set)
Phonon linewidths of bcc Nb (32× 32× 32 k-point set)
Question
Can you think of a situation where at least one of the acousticphonon branches is quadratic for small q, and not linear?